In a 2-year longitudinal study, 66 children were assessed in Grades 3 and 4 with a within-method-and-time design on metacognition (including the computerized Evaluation and Prediction Assessment [EPA2000]) and mathematics. Moreover, half of the children were trained in Grade 3 in the metacognitive skill of predicting their abilities in solving mathematics tasks. All children completed mathematics tests before and after the training took place. Results showed that there was progress in metacognition and in mathematics skills. Moreover, children in the metacognitive group did better than the children in the control group. Conclusions are drawn on the relationship between metacognition and mathematics abilities. (Contains 3 figures and 1 table.)